An Intelligent Optimization and Comprehensive Performance Decision-Making Method for ECC Mix Proportion Based on Machine Learning

Zhichen Liu

doi:10.1016/j.cemconcomp.2026.106532

Return

An Intelligent Optimization and Comprehensive Performance Decision-Making Method for ECC Mix Proportion Based on Machine Learning

2026-02-10

PRE

DOI：10.1016/j.cemconcomp.2026.106532

Original

Original request for help

Save

Abstract

En 中文

Engineered Cementitious Composite (ECC) has been increasingly adopted in frame columns due to its excellent mechanical properties. However, in engineering practice, how to quickly select an appropriate mix design that balances performance and cost-effectiveness remains a challenge. In this study, a machine learning and mechanical performance-driven selection method was established. First, a high-fidelity dataset comprising 240 sets of data with 18 input variables was constructed. Next, hyperparameter tuning of the algorithmic model was performed using random search combined with Fourier optimization, followed by Pearson correlation analysis, prediction accuracy evaluation, and Shapley Additive exPlanations (SHAP) analysis on the dataset. Finally, multiple possible mix designs were generated based on the prediction model and the target concrete compressive strength. These were sequentially processed through a constructed mechanical model, the weight coefficient allocation and normalization calculation process, and a comprehensive performance index model, ultimately obtaining the optimal mix design. A case study was presented to demonstrate the feasibility of the proposed method. In the prediction model, the CatBoost algorithm model developed achieved high prediction accuracy for ECC performance, with R-values of 0.97, 0.95, and 0.91 for compressive strength, tensile strength, and tensile strain, respectively.Factors affecting ECC's mechanical properties mainly include the water-to-binder ratio, cement, silica fume, and fiber fracture elongation, with their average SHAP values over twice those of other inputs. Additionally, a design case validates the rationality of the design method proposed in this study. This study can provide guidance for the broader application of ECC in practical engineering.

Keywords:

Machine Learning

Engineered Cementitious Composite (ECC)

Mix Design Optimization

Mechanical Performance

Shapley Additive exPlanations (SHAP)